|
My point was that there is only so much we can understand. Let me give you a concrete example, one which I have chosen to be easy to understand--the irony!-- : You are a biologist and are given the task of reverting skin senescence in a billionaire client of yours. I'm choosing this example because senescence is a very individual process, with different biological pumps[^1] stopping at different points in time and for different reasons. You can choose to understand how the processes worked together to produce the present system state and skin condition. But that's not your task, your task is to revert it. Understanding seems like a logical first step, but along the way, you (always) discover that these processes involve tens of thousands of interactions between an order of magnitude more of metabolites working at different stages and compartments, and that you can't keep a general intuition of them in your mind[^5], other than the very basic "sh*t breaks". But that's okay. You can always put all of it in a database. Then you only need to remember where the database is, and the dozens of different simulations that are interacting with that database. You will also need to understand the organization of the database, and what the simulations are doing, but there are way less of those and they follow human-made ontologies, sometimes they even come with documentation. If you play with those toys correctly, you will come with an individual intervention for your billionaire that you know will be sound, even if you don't have a comprehensive chain of reasoning of why expressing your customer's variation of 3D9S[^2] 7.5% less will help make his skin better[^3]. In any case, this is an area where there is some vigorous debate[^4] right now. This is somewhat similar to how we don't understand the precise effect of a weight amount trillions in a LLM, but we can still architect, build and profit from the LLM. [^1] That's a name I use for clusters of connected pathways, but the distinction is arbitrary and in this case the clusters were created by a graph clustering algorithm. [^2] https://www.rcsb.org/structure/3D9S [^3] If you are thinking that I should have made this example about cancer: the most frequent cause of cancer is cellular senescence. I couldn't muster the cynicism of making an example about the symptom instead of the cause. But most of my colleagues in search of public funding will. Go figure. [^4] https://direct.mit.edu/posc/article-abstract/31/5/594/115643... [^5] Or, worse, you risk holding to the wrong intuition or understanding. Because we tend to misunderstand complex things much more easily than simple things, you know. |
ie it's not a complex cellular biology thing - just a wear and tear thing, for components that weren't designed to be replaced - ( like adult teeth for example ).
So there might not be an existing biological process you can hijack or reverse - so understanding existing biology might not help you at all.
As to your main point about the complexity of the system. Bottom line biology has evolved to maintain stable patterns - if it was always on a knife edge you'd be dead - so while there might be lots of moving parts the control surface and the state machine has to be much smaller - with the controls being rather forgiving.
As an analogy - you don't need to be a mechanic to be able to drive a car - you can abstract the cars complex mechanics to some very simple high level characteristics - and you can pile those abstractions ( if they don't leak ) on top of each other - so there is a carburettor - you don't need to fully understand how the internals work to understand it's role in the car, but you don't need to know about a carburettor to be able to press the accelerator.